A new quad tree based feature set for recognition of handwritten bangla numerals

Recognition of handwritten Bangla numerals has always been an open problem for researchers. Selection of appropriate preprocessing and feature extraction techniques to achieve maximum recognition accuracy is a challenging problem. In this paper, a new Quad Tree based feature set is introduced for the recognition of handwritten Bangla numeral dataset developed here. On experimentation with the database of 4200 image samples using Support Vector Machine (SVM), the technique yields an average recognition rate of 93.338% evaluated after three-fold cross validation of results. The result is compared with recognition rate obtained from previously established standard dataset using the same feature set.